Resolving Temporary Referential Ambiguity Using Presupposed Content

••

•-

-

-

•

•

•

Dear Author/Editor, Here are the proofs of your chapter as well as the metadata sheets. Metadata

Please carefully proof read the metadata, above all the names and address.In case there were no abstracts for this book submitted with the manuscript, the first 10-15lines of the first paragraph were taken. In case you want to replace these default abstracts,please submit new abstracts with your proof corrections.

Page proofsPlease check the proofs and mark your corrections either by

entering your corrections online oropening the PDF file in Adobe Acrobat and inserting your corrections using the tool"Comment and Markup" orprinting the file and marking corrections on hardcopy. Please mark all corrections indark pen in the text and in the margin at least ¼” (6 mm) from the edge.

You can upload your annotated PDF file or your corrected printout on our ProofingWebsite. In case you are not able to scan the printout , send us the corrected pages viafax.Please note that any changes at this stage are limited to typographical errors and seriouserrors of fact.If the figures were converted to black and white, please check that the quality of suchfigures is sufficient and that all references to color in any text discussing the figures ischanged accordingly. If the quality of some figures is judged to be insufficient, please sendan improved grayscale figure.

Metadata of the chapter that will be visualized onlineBook Title Experimental Perspectives on Presuppositions

Chapter Title Resolving Temporary Referential Ambiguity Using PresupposedContent

Copyright Springer International Publishing Switzerland 2015

Corresponding Author [Aff 1] Prefix

Family name Romoli

Particle

Given name Jacopo

Suffix

Division School of Communication

Organization University of Ulster

Address Belfast, United Kingdom

Email [email protected]

Author [Aff 2] Prefix

Family name Romoli

Particle

Given name Jacopo

Suffix

Division Department of Cognitive Science

Organization Macquarie University

Address Sydney, Australia

Author Prefix

Family name Khan

Particle

Given name Manizeh

Suffix

Division Department of Psychology

Organization Harvard University

Address Cambridge, USA



Family name Sudo

Particle

Given name Yasutada

Suffix

Division Linguistics

Organization UCL

Address London, United Kingdom



Family name Sudo

Particle

Given name Yasutada

Suffix

Division Linguistics and Philosophy department

Organization MIT


Author Prefix

Family name Snedeker

Particle

Given name Jesse

Suffix

Division Department of Psychology

Organization Harvard University



Abstract We present the results of two visual-world experiments investigatingwhether the presupposition of ‘also’ is used to predict upcominglinguistic material during sentence comprehension. We comparepredictions generated by ‘also’ to predictions from a parallel inferencegenerated by ‘only’ (i.e., that the upcoming material will be unique).The results show that adults do use the presupposition of ‘also’incrementally in online sentence comprehension and they can do sowithin 200 to 500 ms of the onset of the presuppositional trigger.Furthermore, they use it regardless of whether contextual support isexplicit or implicit. On the other hand, we did not observe effects ofthe inference generated by ‘only’ at any point during the sentence,even though this information was used in an offline task.

Keywords Presuppositions - Entailments - Processing - Visual-world -Semantics - Pragmatics - Psycholinguistics

Unc

orre

cted

Proo

f

Book ID: 316680_1_En ChapterID: 3 Dispatch Date: 06-07-2014 Proof No: 1

Resolving Temporary Referential AmbiguityUsing Presupposed Content

Jacopo Romoli, Manizeh Khan, Yasutada Sudo and Jesse Snedeker

Abstract We present the results of two visual-world experiments investigating1

whether the presupposition of ‘also’ is used to predict upcoming linguistic mate-2

rial during sentence comprehension. We compare predictions generated by ‘also’ to3

predictions from a parallel inference generated by ‘only’ (i.e., that the upcoming4

material will be unique). The results show that adults do use the presupposition of5

‘also’ incrementally in online sentence comprehension and they can do so within6

200 to 500 ms of the onset of the presuppositional trigger. Furthermore, they use it7

regardless of whether contextual support is explicit or implicit. On the other hand,8

we did not observe effects of the inference generated by ‘only’ at any point during9

the sentence, even though this information was used in an offline task.10

Keywords Presuppositions · Entailments · Processing · Visual-world · Semantics ·11

Pragmatics · Psycholinguistics12

J. Romoli (�)School of Communication, University of Ulster, Belfast, United Kingdome-mail: [email protected]

Department of Cognitive Science, Macquarie University, Sydney, Australiae-mail: [email protected]

M. KhanDepartment of Psychology, Harvard University, Cambridge, USAe-mail: [email protected]

Y. SudoLinguistics, UCL, London, United Kingdome-mail: [email protected]

Linguistics and Philosophy department, MIT, Cambridge, USAe-mail: [email protected]

J. SnedekerDepartment of Psychology, Harvard University, Cambridge, USAe-mail: [email protected]

1F. Schwarz (ed.), Experimental Perspectives on Presuppositions,Studies in Theoretical Psycholinguistics 45, DOI 10.1007/978-3-319-07980-6_3,© Springer International Publishing Switzerland 2015

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


2 J. Romoli et al.

1 Introduction13

Presuppositions are a class of inferences that we draw from utterances and are gen-14

erally characterized along two dimensions: their discourse role and their behavior15

in embeddings.1 To illustrate, consider a sentence like (1), from which we typically16

draw the two conclusions in (2a) and (2b).17

(1) Fred stopped smoking.18

(2) a. Fred doesn’t smoke.19

b. Fred used to smoke.20

In the literature, inferences like (2b) are referred to as ‘presuppositions,’ while those21

like (2a) as ‘entailments.’ These two inferences are considered to be different for22

two reasons. First, they intuitively play different conversational roles: (2a) is new23

information added to the context, whereas (2b) is typically assumed to convey infor-24

mation that is given at the point of utterance of (1). We can refine this intuition by25

explicitly adding each of these inferences before the statement itself. When we add26

the presupposition to the statement, the result (shown in (3b)) is a natural discourse.27

In contrast, when the other inference is placed before the statement, the result (shown28

in (3a)) is unnatural. The standard way to account for the oddness of (3a) is to appeal29

to a condition that requires a speaker to not assert anything that is redundant in its30

context of utterance (see Stalnaker 1978 and much subsequent work). Crucially, this31

condition does not apply to presuppositions, resulting in the contrast between (3b)32

and (3a).233

(3) a. Fred doesn’t smoke and he now stopped.34

b. Fred used to smoke and he now stopped.35

The second distinctive property of presuppositions is their behavior in complex sen-36

tences. Consider (4a)–(4d) below in which the statement (1) is embedded in different37

complex sentences.38

(4) a. Fred didn’t stop smoking.39

b. Did Fred stop smoking?40

c. If Fred stops smoking, Lisa will be happy.41

d. It’s possible that Fred stopped smoking.42

All of these sentences still generate the inference in (2b) but they do not generate the43

conclusion in (2a). In other words, the presuppositions of sentences like (1), unlike44

entailments, appear to be ‘inherited’ by most of the complex sentences containing45

them. This pattern, traditionally called ‘projection behaviour’, is characteristic of46

presuppositions and it is generally used as the primary diagnostic for distinguishing47

1 For an introduction to presuppositions see Chierchia and McConnell-Ginet 2000 and Beaver andGeurts (2011).2 In fact, one approach, stemming again from the work of Stalnaker (1974, 1978), is to think thatpresuppositions not only can but have to be redundant in the context of utterance.

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


Resolving Temporary Referential Ambiguity Using Presupposed Content 3

presuppositions from other inferences. Accounting for how presuppositions project48

is a central topic in semantics and pragmatics (Karttunen 1974; Stalnaker 1974;49

Gazdar 1979; Heim 1983; van der Sandt 1992 and Beaver 2001. For some recent50

proposals see Schlenker 2008 and Schlenker 2009).51

While presuppositions have been studied extensively, we still known little about52

how they are used during comprehension, as the sentence unfolds. This gap is impor-53

tant to fill because presuppositions carry information that could be used incrementally54

to guide interpretation during language processing. By tracing how this informa-55

tion becomes available over time, we could learn more about the interplay between56

pragmatics and compositional semantics during language comprehension.57

1.1 The Processing of Presupposed Content58

Recently, people have started looking at the question of how presuppositions are59

processed during comprehension (Kim 2007; Schwarz 2007; Schwarz and Tiemann60

2013a, 2013b). Three broad questions have guided this work. First, is the presup-61

posed content of an utterance available as quickly as its assertive content? Or do62

new presuppositions systematically lag behind assertions, in a manner that parallels63

scalar implicatures (Bott and Noveck 2004; Huang and Snedeker 2009 among many64

others)? Second, is there a processing cost to presupposition violation? Observing65

such a cost would also inform the first question (by placing an upper bound on the66

time by which presupposition was calculated). Third, once presuppositions are avail-67

able, how do they affect sentence processing? Are they used to resolve ambiguities68

at other levels or make predictions about upcoming referents? Again, data on this69

question would also constrain answers to the first.70

Kim (2007) explored how presupposed content and asserted content are accessed71

during a sentence verification task. Participants were shown various visual displays72

and asked to judge whether sentences like (5) were true or false. Their task was to73

press a button corresponding to ‘yes’ if the sentence accurately described the visual74

context and ‘no’ if it did not.75

(5) Only the girls have books.76

Kim (2007) adopted the analysis of ‘only’ in which a sentence like (5) asserts (6a)77

and presupposes (6b) (see Beaver and Clark 2009 and references therein).78

(6) a. No people other than the girls have books.79

b. The girls have books.80

There were two types of critical trials which varied in terms of the picture that was81

paired with the utterance. On the false assertion trials, (5) was matched with a pic-82

ture in which two girls out of eight characters had books (i.e., the presupposition83

was satisfied) and some of the other six characters also had books (i.e., the assertion84

was false). On the presupposition violation trials, (5) was matched with a picture in85

which the two girls didn’t have books and the other characters did not have them86

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


4 J. Romoli et al.

either (i.e., presupposition false, assertion true). She found that participants were87

faster to reject the false assertion trials than they were to reject the presupposition88

violations. On the basis of these results, Kim (2007) concludes that the information89

conveyed by presuppositions and the information conveyed by assertions are used90

differently in sentence comprehension. Comprehenders do not have to verify pre-91

suppositions before evaluating the truth of a sentence. Instead, they assume that the92

presuppositions are true and only check them afterwards.93

Schwarz (2007) used a reading time paradigm to explore the comprehension94

of sentences with presupposition triggers in both German and English. In one of95

the experiments, participants had to read sentences like (7) and (8), which differ96

in the content of the relative clauses. Crucially the relative clause (7) satisfies the97

presupposition of ‘also’ (i.e., that there is some other relevant person to whom the98

congressman wrote to), but (8) does not.99

(7) The congressman, who wrote to John, had also written to the mayor to schedule100

a meeting for the fundraiser.101

(8) The congressman, who John wrote to, had also written to the mayor to schedule102

a meeting for the fundraiser.103

The stimuli were presented in a phrase-by-phrase self-paced reading paradigm in104

which multiple words were presented at once. The experimental region included105

the presuppositional trigger and several subsequent words (e.g., had also written to106

the mayor). The participants were slower to read the experimental region when the107

relative clause did not satisfy the presupposition associated with ‘also’ (8) than when108

it did (7).109

This finding demonstrates that there is a processing cost in cases where there is also110

presupposition failure. We believe that there are at least two possible explanations for111

this processing cost. First, this cost could be a direct consequence of presupposition112

failure: participants could slow down in (8) because they fail to find any prior event113

of the relevant kind (one with the congressman writing someone) and thus cannot114

integrate the presupposition triggered by ‘also’. In this case, the cost should only115

come after encountering the main verb in the active voice (had also written), since116

that information is needed to determine that the relative clause does not satisfy the117

presupposition (e.g., see (9)).3 Alternatively, the cost could reflect the participants’118

predictions about the form or meaning of the verb, given the relative clause and119

presuppositional trigger. Specifically, in (7) the participant should expect the verb120

that is used (written), while in (8) they might expect a verb with a different argument121

structure (again see (9)).122

3 Schwarz seems to interpret his online data as a reflection of presupposition failure, rather thanprediction (see p. 402). However, his work also provides evidence that presuppositions can be usedto predict and/or revise the thematic roles in an event. Specifically, in an offline study in German,he finds that readers tend to interpret a relative clause with ambiguous case marking in a mannerthat would allow to satisfy the presupposition of ‘auch’ (also).

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



(9) The congressman who John wrote to had also received a letter from the mayor123

to schedule a meeting for the fundraiser.124

In this paper, we report the results of two experiments that investigate the time course125

of presupposition calculation by looking for effects of a presupposition on the inter-126

pretation of an upcoming noun. Like Schwarz (2007), we focus on the presupposition127

of ‘also’. Unlike Schwarz (2007), we are not concerned with the effect of presuppo-128

sition failure on processing. Instead we investigate how early participants are able129

to use the information associated with presuppositions to determine the referents of130

upcoming nouns. In addition, we compare this effect to a parallel inference based on131

the entailments associated with ‘only’. Our focus differs from that of Kim’s (2007)132

study, as we are not looking at how presupposed versus asserted content is used in133

verification but whether presupposed content is used to predict an upcoming word.4134

1.2 Experimental Paradigm135

In the experiments reported in this paper, we use the visual world paradigm, because136

it provides a sensitive and time-locked measures of language comprehension. Prior137

studies have successfully used this method to study a broad range of linguistic pro-138

cesses including: word recognition (e.g., Allopenna et al. 1998), syntactic ambiguity139

resolution (e.g., Tanenhaus, Spivey-Knowlton, Eberhard, and Sedivy 1995) and the140

calcuation of scalar implicatures (e.g., Huang and Snedeker 2009). The stimuli and141

task used in Experiment 1 are closely modeled on a series of experiments conduced142

by Kim et al. (2008).143

Kim and colleagues presented participants with pairs of sentences like (10) and144

(11). As mentioned above, a sentence like (11) is generally assumed to presuppose145

(12a) and assert something along the lines of (12b).146

(10) Mark has some candies and some shoes.147

(11) Jane only has some candies.148

(12) a. Jane has candies.149

b. Jane has no relevant things other than candies.150

Kim and colleagues were interested in how listeners restrict the assertion of a sen-151

tence like (11). Can context be used to determine the set of relevant things as the152

4 There are two other recent studies on presupposition processing that are not directly relevant to thequestion posed here (Schwarz and Tiemann 2013a, 2013b and Chemla and Bott 2013). Schwarz andTiemann look at whether the position of the trigger in a complex sentence influences the processingof presuppositions. On the basis of their results they argue that there is a processing cost associatedto the ‘length of projection’, that is the distance between the position of the trigger and the clauseassociated to the information satisfying the presupposition appears. Chemla and Bott, on the otherhand, concentrates on whether the accommodation of presupposition is more costly when it occursabove or below negation (see also Romoli and Schwarz (this volume)).

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


6 J. Romoli et al.

Fig. 1 An example of adisplay from Kim et al 2009.The relevant sentences arelisted in (13). Illustrationcourtesy of Christina Kim

sentence unfolds and what kinds of contextual cues are relevant? They assessed this153

by measuring the participants’ ability to predict the upcoming noun on the basis of154

the context.155

Participants are presented with spoken instructions, asking them to choose among156

different objects within a visual reference world and their eye movements are mea-157

sured. Each trial contained a context sentence (10) and one of four target sentences158

(13a-13d) below.159

At the onset of the target sentence, a grid of four images appeared (Fig. 1) which160

consisted of the target object (for (11) ‘candies’), a cohort competitor (for (11)161

‘candles’) and two distractors (for (11) ‘anchors’ and ‘shoes’). The competitor is162

crucial in this design, because it insures that the identity of the noun is unclear even163

after the word begins providing a longer window in which predictions generated by164

the context and the structure of the sentence can guide interpretation.165

The target sentences were constructed by manipulating two variables: whether166

‘only’was present and whether the target noun was mentioned in the context sentence167

(‘old’ versus ‘new’). This resulted in four conditions, as summarised in (13).168

(13) a. Jane only has some candies. (only/old)169

b. Jane has some candies. (control/old)170

c. Jane only has some candles. (only/new)171

d. Jane has some candles. (control/new)172

If context is used to restrict the relevant set, then participants who heard the sentences173

with ‘only’ should assume that the upcoming noun will be one of the previously174

mentioned items (e.g., candies). Thus they should be faster to reach the target in175

sentences like (13a) than in sentences like (13b). This systematic preference for the176

previously mentioned object should be absent or reduced in sentences without ‘only’177

resulting in an interaction between the two variables. This is precisely what Kim and178

colleagues found, leading them to conclude that context is used to rapidly restrict179

the comparison class. We return to these findings in the general discussion section180

(to explore their relevance to our ‘only’ control condition).181

Our study employs a similar design to explore a different question. Like Kim182

and colleagues, we provided participants with: a context sentence that introduced183

two objects, a target sentence referring to one of them, and a display that contained184

two cohort competitors. However, our goal was to use this paradigm to explore185

how the presupposition of ‘also’ can be used to predict an upcoming referent. Our186

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



critical items consisted of series of three sentences (14). The first (14a) established187

a set of two characters, making it natural to draw comparisons between them. The188

second sentence (14b) described what one character had. The critical third sentence189

(14c) contained the presupposition trigger. This sentence was intended to have the190

interpretation where ‘also’ is associated with the subject. To achieve this the utter-191

ances were produced with stress on the subject and a prosodic break after ‘also’ (see192

Appendix B).193

(14) a. Mark and Jane are friends (introduction)194

b. Mark has some candies and some shoes (context)195

c. JANE, also, has some candies. (target)196

We are assuming that a sentence like (14c) presupposes something along the lines of197

(15a) and asserts (15b) (see Kripke 2009 and Heim 1992 among others).198

(15) a. A relevant individual in the context other than Jane has some candies.199

b. Jane has some candies.200

In both of the following experiments, we manipulated whether the intended target201

referent was mentioned or not in the context sentence (‘old’ versus ‘new’) whether202

there was a focus particle in the sentence which could allow participants to predict203

the discourse status of the referent before encountering it (‘disambiguated’ versus204

‘control’). For old referents the disambiguating focus particle was ‘also’, for new205

referents it was ‘only.’ This resulted in the following four conditions in (16).206

(16) a. Jane, also, has some candies. [also/old]207

b. Jane has got some candies. [control/old]208

c. Only Jane has some candles. [only/new]209

d. Jane has got some candles. [control/new]210

The logic of the design is as follows. In the control conditions, participants have211

no basis on which to predict the final object, and thus they should split their gaze212

between the two cohort members from the onset of the final noun until phonologi-213

cal disambiguation, after which they should quickly converge on the target object.214

In contrast, in the also/old condition, the presuppositional trigger provides infor-215

mation that could allow participants to infer the target nouns before phonological216

disambiguation. Specifically, if participants are able to rapidly determine the presup-217

position of also they could potentially infer that the target is one of the previously218

mentioned items (candies or shoes) anytime after the trigger, allowing them to de-219

termine which noun it is (candies) after hearing the first phoneme. Thus we expect220

that folks will look at the correct target more often and more quickly in the also/old221

condition than in the control/old condition.222

The only/new condition was included to ensure a balanced design. In this case223

the focus particle favors the novel referent: if participants incrementally recover the224

assertion in the ‘only’sentences they should recognize that the upcoming object must225

be something that only Jane has, and thus cannot be one of the previously mentioned226

objects. This should lead them to resolve the referential ambiguity in favor of the227

new target before phonological disambiguation (Fig. 2).228

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


8 J. Romoli et al.

Fig. 2 An example of adisplay from Experiment 1.The relevant sentences arelisted in (13).llustration–author’s own

2 Experiment 1229

2.1 Method230

2.1.1 Participants231

Thirty two participants were recruited from the Harvard community, including un-232

dergraduate students. They received either course credit or $ 5 for their participation.233

All participants were native speakers of English. One participant was excluded due234

to low accuracy with the fillers.235

2.1.2 Material236

Each item consisted of: a set of four pictures, an introduction sentence (17), a237

context sentence (18), and a critical sentence (13). The four pictures included two238

pictures which could readily be described by nouns sharing an onset (e.g., candies and239

candles). The introduction sentence presented two characters. The context sentence240

indicated that one character had two items: one of the cohort members (candies)241

and one of the non-cohort members (watches). As we noted above, there were four242

different variants of the target sentence.243

(17) Introduction sentence244

Mark and Jane are friends245

(18) Context sentence246

Mark has some candies and some watches.247

(19) a. Jane, also, has some candies. [also/old]248

b. Jane has got some candies. [control/old]249

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



c. Only Jane has some candles. [only/new]250

d. Jane has got some candles. [control/new]251

Both the presence of the focus particle and the discourse status of the final noun were252

manipulated within subjects and within items. It is important to note that the ‘only’253

sentences predict a target noun that was not in the context sentence, in contrast to254

the ‘also’ sentences that presuppose that the target noun was in the context sentence.255

The predictions for these sentences were confirmed in a norming study conducted256

on Amazon Mechanical Turk (Appendix A). As mentioned, we also conducted a257

norming task to ensure that the intonation used in the ‘also’ sentences evoked the258

expected interpretation (Appendix B). In addition, we wanted to control which of259

the two cohort members appeared in the context sentence. To counterbalance all260

three of these variables, we created eight versions of each of the 32 base items. We261

then created eight lists such that each base item appeared only once on a given list262

and in all eight cells across the lists. Thus each participant heard each of the four263

target sentence types eight times over the course of the experiment. Eight filler trials264

were included. Two fillers appeared at the beginning to make sure participants were265

familiar with the task before the first experimental trial. The remaining six filler trials266

were interspersed throughout the experiment. Trial order was pseudo-randomised,267

with each participant seeing base items in the same order. The positions of the268

different kinds of objects (target, competitor and distractors) in the visual display269

were counterbalanced across trials.270

2.1.3 Procedure271

Participants were seated at a comfortable distance from the screen of a Tobii T-272

60 eye-tracker. The auditory stimuli were played by the computer through external273

speakers. First, participants heard the introduction sentence and context sentence.274

Then a fixation point appeared in the middle of the screen. After the participant had275

fixated on this point for 500 ms, the target sentence began. This was done to ensure276

that participants were always looking in the same place at the beginning of the critical277

sentence. At the onset of the target sentence, the images appeared and participants’278

fixations were measured. The participants were told that their task was to pick the279

last object mentioned in the target sentence by clicking the relevant picture. Once280

the participant did this the trial ended and the next trial began.281

2.1.4 Results282

We analyzed the log-odds of the proportion of fixations to the old cohort item (the283

competitor that had been mentioned in the context sentence) versus the new cohort284

item (Fig. 3). Prior to the disambiguation point of the noun, the control sentences285

are compatible with either image. If participants incrementally incorporate the pre-286

supposed content of ‘also’ during sentence processing, then we should expect to287

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


10 J. Romoli et al.

see more looks to the old cohort item in sentences with ‘also’ compared to control288

sentences.289

Also Our primary analyses compared the also-old trials to the control-old trials. We290

conducted linear mixed-effects regressions, with the maximally appropriate random291

effects structure, looking at the log-odds of proportion of looks to the old cohort. We292

estimated p-values using the pnorm function in R. We examined four time windows:293

the first of these time windows coincided with the noun (offset by 200 ms), while294

the other three spanned the 900 ms before the noun (− 700 to − 400, − 400 to295

− 100, and −100 to 200). On average, the onset of the word ‘also’ occurred 806 ms296

before the onset of the target noun, so these time windows would cover saccades297

programmed at the very onset of ‘also’(early pre-noun), as well as the period in which298

the presuppositional trigger could potentially generate predictions about upcoming299

referents (the mid and late pre-noun windows).300

During the noun window, participants were significantly more likely to look at301

the old cohort in also-old sentences compared to control-old sentences (t = 4.20,302

p < 0.001). This pattern was already evident in the late pre-noun window which303

covered the 300 ms immediately preceding the noun (t = 3.13, p < 0.002). There304

were no significant differences between the ‘also’ and the control-old sentences in305

early and mid pre-noun time windows.306

Only We were also interested in whether participants used the information from307

‘only’to anticipate unique referent. That is, did participants look at the old cohort less308

in the ‘only’ sentences compared to the control-new sentences? We conducted linear309

mixed-effects regressions, parallel to those above. Again we used the maximally310

appropriate random effects structure and looked at the log-odds of proportion of311

looks to the old cohort during the four time windows described earlier. There were312

no significant differences between looks to the old cohort in the ‘only’ sentences and313

the control-new sentences (ts < 1, ps > 0.3). This null effect is surprising given that314

participants correctly predicted the target noun on the basis of ‘only’ in an offline315

task (see Appendix A).316

2.1.5 Discussion317

Experiment 1 provides a clear answer to our original questions about the processing318

of presuppositions during language comprehension. First, we found that presupposed319

content can be used to guide lexical (or referential) predictions. In the ‘also’condition,320

participants showed a preference to look at the previously-mentioned cohort member.321

This preference began well before noun onset, resulting in a robust difference between322

the ‘also’ sentences and their controls in the late pre-noun time window (− 100 ms to323

200 ms unshifted). Second, the timing of effect places an upper-bound on the point324

at which the presupposition is calculated. The effect of the presupposition became325

robust in a time window that began about 700 ms after the onset of ‘also’. If we326

make the standard assumption that it takes a minimum of 200 ms for information327

in the speech stream to affect saccades (Matin et al. 1993; Allopenna et al. 1998),328

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



-1.5

-1

-0.5

0

0.5

1

1.5

2

EarlyPre-Noun

MiddlePre-Noun

LatePre-Noun

Noun

Log

-odd

s of p

ropo

rtio

n of

look

s to

old

item

Time region

"Also" - OldControl - Old"Only" - NewControl - New

Fig. 3 Experiment 1: gaze data

then this suggests that it takes 500 ms or less, from the onset of the trigger, for the329

presupposition to be calculated.330

However, this experiment also provided an unexpected surprise. To ensure balance331

in our stimuli, we had included a condition in which ‘only’had subject scope (20). As332

we noted earlier, it is standard to assume that these utterances have the presupposition333

given in (21a) and express the assertion given in (21b).334

(20) Only Jane has some candles.335

(21) a. Jane has some candles.336

b. No other relevant people have candles.337

Thus we expected that the ‘only’ sentences would allow the participants to rule out338

the previously mentioned items as potential referents for the final noun, because they339

would make the assertion false (since the other character did have those items). But340

curiously, we failed to find smallest hint of this effect in the eye-movement data,341

even though an offline norming study confirmed that participants were sensitive to342

this constraint (Appendix A).343

Experiment 2 had three goals. First, we wanted to replicate our critical finding344

that presuppositions can be used to predict upcoming words or referents. Second,345

we wanted to extend these results by exploring whether more indirect contextual346

support would be sufficient to satisfy the presupposition and guide online processing.347

Specifically, in Experiment 1, the presupposed content had been explicitly mentioned348

(‘Michael has some candles’). In Experiment 2, we compare explicit mention to349

implicit mention (‘Look at what Michael has’).350

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


12 J. Romoli et al.

Finally, we wanted to create a discourse context that would be more appropriate351

for the ‘only’ sentences, to determine whether this would allow people to use this352

assertion to predict the upcoming noun in real time. We reasoned that sentences like353

(20) are felicitous only in contexts in which there is some relevant set of people that354

the subject (‘Jane’) is being contrasted with. In Experiment 1, the discourse included355

just two people. Since we had already stated what the other person had, the assertion356

in the ‘only’ sentence was actually no more informative than the presupposition.357

Consequently, the control-new sentence was arguably a more felicitous way to ex-358

press this idea. In Experiment 2, there are a total of three people in the context and359

the task centers around distinguishing the person in the target sentence from another360

person.361

3 Experiment 2362

3.1 Method363

3.1.1 Participants364

Thirty two participants were recruited from the Harvard community, including un-365

dergraduate students. Half of the participants were assigned to the explict context366

condition and half to the implicit context condition. They received either course367

credits or $ 5 for their participation. None of them had participated in Experiment368

1. All participants were native speakers of English. Eight participants were excluded369

(2 for low response accuracy, 1 for software error, and 5 for excessive track loss).370

3.1.2 Procedure and Material371

Three changes were made to the materials from Experiment 1. First, two between372

participant conditions were created. In the explicit context condition (22), the context373

sentence (22a) overtly mentions the critical objects, just like the context sentence in374

Experiment 1.375

(21) Explicit Context376

a. Context Sentence: Michael has got candies and watches.377

b. ‘Also’Target: Sarah also has some candies.378

c. ‘Only’Target: Only Sarah has some candles.379

In the implicit context condition (23), the context sentence (23a) directs attention to380

these objects but does not mention them by name.381

(22) Implicit Context382

a. Context Sentence: Look at what Michael has!383

b. ‘Also’Target: Sarah also has some candies.384

c. ‘Only’Target: Only Sarah has some candles.385

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



Fig. 4 An example of thevisual context used inExperiment 2. The relevantsentences are in (22) and (23).llustration—author’s own

Second, we created new visual displays which included three new characters (Fig. 4).386

One character matched the gender of the person in the context sentence (the man at387

the top). This character had the two objects mentioned in the context sentence and388

was always placed on the top half of the screen in the center. The two other characters389

matched the gender of the person in the target sentence. One of these characters was390

the expected referent for the ‘also’ sentences because s/he had the old cohort object391

as well (the woman with candies on the right). The other character was the expected392

referent for the ‘only’ sentences because she had a unique object (the woman on the393

left with the candles). These characters appeared on the bottom half of the screen,394

with their relative position counterbalanced across trials.395

Third, because the characters were now visible on the slide, we removed the396

introductory sentence which had linked the two protagonists.397

The procedure was also modified. Participants were told that their task was to click398

on the person mentioned in each sentence. At the beginning of each trial, the visual399

display appeared, followed by the context sentence. After the participant clicked400

on the character mentioned in the context sentence, the target sentence was played.401

The trial ended when the participant clicked on the character mentioned in the target402

sentence. As in Experiment 1, each participant heard four different kinds of target403

sentences (also-old, control-old, only-new, and control-new) and there were eight404

trials per participant, in each condition.405

3.1.3 Results and Discussion406

As with Experiment 1, we analyzed the log-odds of the proportion of fixations to407

the old cohort item. We defined looks to the old cohort item as looks to the quadrant408

that contained the new person who had the old cohort object (lower right quadrant in409

Fig. 4). Similarly, we defined looks to the new cohort item as looks to the quadrant410

with the person who has the new cohort object (lower left quadrant in Fig. 4). We411

analyzed the also-old sentences in comparison to the control-old condition, and412

the only-new sentences in comparison with the control-new sentences. We were413

interested both in main effects of the markers and potential interactions between the414

marker and the context condition.415

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


14 J. Romoli et al.

-1

-0.5

0

0.5

1

1.5

EarlyPre-Noun

MiddlePre-Noun

LatePre-Noun

Noun

Log

-odd

s of p

ropo

rtio

n of

look

s to

old

item

Time region

Explicit Context -"Also"Explicit Context -ControlImplicit Context -"Also"Implicit Context -Control

Fig. 5 Experiment 2: comparison of the also and control-old sentences

Also We conducted linear mixed-effects regressions with sentence type (also vs.416

control) and context condition (explicit vs. implicit) as predictors, with the maximally417

appropriate random effects structure. We estimated p-values using the pnorm function418

in R. We examined the same four time windows used in Experiment 1: the noun419

window and the three 300 ms timebins before the noun (early, mid, and late pre-420

noun). As in experiment one, on average, the onset of the word ‘also’ occurred421

around 800 ms before the onset of the target noun.422

During the noun window, there was a main effect of sentence type (t = 5.26, p423

< 0.001) (Fig. 5). When the sentence contained ‘also’, participants shifted their gaze424

to the old cohort item. Critically, this effect was also significant in both the mid425

and late pre-noun windows (both t’s > 3, p’s < 0.002), but not in the early pre-noun426

time window. This pattern indicates that the presupposed information was available427

and able to guide reference resolution within 200 to 500 ms after encountering the428

trigger (allowing 200 ms to program a saccade). There was no main effects of or429

interactions with the context condition, suggesting that the presupposed content was430

integrated incrementally regardless of whether the anaphoric antecedent of ‘also’was431

explicitly mentioned. To address this question more directly, we conducted separate432

analyses of each context condition. We found a reliable difference between the also433

and control-old sentences in both the explicit and implicit conditions (t = 3.60, p <434

0.001 for the explicit context condition; t = 3.68, p < 0.001 for the implicit context435

condition).436

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

EarlyPre-Noun

MiddlePre-Noun

LatePre-Noun

Noun

Log

-odd

s of p

ropo

rtio

n of

look

s to

old

item

Time region

Explicit Context -"Only"Explicit Context -ControlImplicit Context -"Only"Implicit Context -Control

Fig. 6 Experiment 2: Comparison of only and control-new sentences

Only We conducted a parallel analysis of the ‘only’and control-new sentences using437

the same variables and approach. If participants were using the information from438

‘only’ to anticipate a unique referent, then we would expect them to shift away from439

the old cohort item well before the noun onset, resulting in a main effect of sentence440

type. This is not what we observed (Fig. 6). We found no significant differences441

between the sentence types in any of the time windows (all t’s < 0.9, p’s > .3).442

In addition, there was no effect of context condition and no interactions between443

context and sentence type (all p’s > .3).444

In sum, as in Experiments 1, hearing ‘also’ led to anticipatory eye movements445

to the target image but hearing ‘only’ did not, despite our attempt to create a more446

felicitous context for the use of ‘only’. Further, participants used the information from447

the presupposition of ‘also’ regardless of whether the discourse context explicitly448

mentioned the repeated item or only implicitly referred to it.449

4 General Discussion450

The primary goal of these experiments was to explore the online processing of pre-451

supposition, and we were largely successful in doing this. Along the way, however,452

we encountered a strange failure in the interpretation of ‘only’. We discuss these two453

findings in turn.454

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


16 J. Romoli et al.

4.1 Presuppositions and Incremental Interpretation455

We had three specific questions about the processing of presuppositions, which we456

were able to answer in these experiments.457

First, can presupposed information be used to make predictions about upcoming458

lexical items during language comprehension? In other words, do presuppositions459

feed back into language comprehension creating expectations about how a sentence460

will end? In both Experiment 1 and Experiment 2, we found that the presupposition461

lead participants to identify the correct referent well before they had any information462

about the critical word.463

Second, how soon after the trigger is the presupposition calculated? As we noted464

earlier, Schwarz (2007) found that readers slow down when they encounter a clause465

with a presuppositional violation, demonstrating that presuppositions are calculated466

as we read. Schwarz’s data place a loose upper bound on the timing of this process:467

on average the presupposition must have been available at some point within 1500 ms468

after the clause was presented. Our findings refine that estimate. In Experiment 2, the469

difference between the also-old and control-old conditions was robust in the middle470

pre-noun time window, a period roughly 400 to 700 ms after the onset of the presup-471

positional trigger. If we make the standard assumption that it takes about 200 ms to472

launch a saccade in response to a phonological cue (Matin et al. 1993, Allopenna et473

al. 1998), then we can conclude that the presupposition was available within 200 to474

500 ms of word onset. Put another way, it appears that the presupposition generated475

by ‘also’ can sometimes be calculated before the word itself is finished.476

At first glance, our findings might seem hard to reconcile with Kim’s (2007)477

picture verification study. Recall that she found that a sentence with ‘only’ was478

faster to evaluate when the assertion was false and the presupposition was true,479

than when the presupposition was false but the assertion was true. She concluded480

that the presupposition of a sentence is checked after the asserted content has been481

checked. That could be taken to suggest that presuppositions are not calculated until482

a late stage in verification. If that was the intended interpretation, then it would be483

incompatible with our findings (as well as with Schwarz’s 2007). However, we see no484

reason to interpret the finding in that way. It is entirely possible that folks calculate485

presuppositions rapidly and incrementally but do not check these presuppositions486

immediately when verifying a statement against a single stable context. In fact, that487

might be a very smart thing to do: if presuppositions can generally be assumed to488

be true, then we ought to put highest priority on verifying the more contentious489

assertions.490

The final question that we explored was: What forms of context can be accessed491

to satisfy a presupposition during online processing? The predictive inference that492

we saw in these experiments was based on the assumption that the presupposition493

of ‘also’ would be satisfied if the direct object of the target utterance referred to494

something that someone else in the discourse context also had in their possession.495

The results of Experiment 1 demonstrate that participants can quickly access ma-496

terial from the sentence immediately before the target sentence to find a suitable497

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



prior referent. Experiment 2 refines this in several ways. First and most obviously,498

it shows that participants will infer that an indirect reference can satisfy a presuppo-499

sition even when no direct reference is available. This is interesting in part because500

the presuppositional trigger that we used (‘also’) is typically argued to be more diffi-501

cult to accommodate than soft triggers such as verbs like win or stop (Simons 2001,502

Abusch 2010, Romoli 2012, Romoli (to appear) among others). Our present findings503

do not challenge theories that propose that the presupposition of ‘also’ has pronom-504

inal characteristics—our indirect contexts would be sufficient to ground a pronoun505

(24). But they point the way toward manipulations which could explore this more506

thoroughly. For example, contexts like the one in (25) do not support the use of a507

pronoun but might allow for the predictive use of presuppositions (26).508

(23) Look at what Michael has! It is good to eat.509

(24) Look at Michael.? It is good to eat.510

(25) Look at Michael. Jane, also, has some candies.511

The parallelism between the implicit and explicit context conditions is important512

because it suggest that, in this paradigm at least, there is no apparent cost to indirect-513

ness. Not only did participants use the indirect context to predictively, they did so as514

rapidly and efficiently as participants in the explicit context conditions. This suggests515

that the processes involved operate over discourse entities, rather than over lexical516

items. Perhaps this isn’t surprising, since presuppositions are typically characterized517

as discourse level expectations, but it is reassuring.518

4.2 A Surprising Failure519

In both Experiment 1 and Experiment 2, we found absolutely no evidence that par-520

ticipants could use the inference associated with ‘only’ to predict that the upcoming521

object would be new to the discourse. We find this surprising for three reasons.522

First, our intuition that this inference is accessible was validated in an offline523

study (see Appendix A). When given sufficient time, folks realize that only the new524

cohort item can complete the sentences with ‘only’.525

Second, in our task at least, the ‘only’ inference seems roughly comparable in526

complexity and constraint to inference in the ‘also’ condition. In both cases, partic-527

ipants must track the referents mentioned in the context sentence and then use this528

information to pick a possible referent in the target sentence. In Experiment 1, both529

of the critical words could potentially be used to focus in on two out of four referents530

prior noun onset. In Experiment 2, both the critical words could potentially allow531

the listener to focus on a single correct referent prior to the noun. In both cases, the532

critical word is associated with the subject but is being used to make an inference533

about the direct object. In fact, the differences that exist seem to favor ‘only’: the534

critical cue comes earlier in the sentence and the focused element is disambiguated535

by word order alone.536

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


18 J. Romoli et al.

Finally, at first glance, our results seem to conflict with those of Kim and col-537

leagues (2008) who found that participants made rapid use of ‘only in sentences like538

‘Jane only has some candy.’ Critically, these studies used essentially the same design539

as we used in Experiment 1. There are two critical differences between the Kim study540

and the present ‘only’ condition, which suggest two hypotheses about why adults are541

unable to make this inference in real time.542

One difference is that in the Kim study ‘only’ associates with the VP, while in our543

study it associates with the subject. Crain and colleagues have found that children544

have more difficulty interpreting subject ‘only’ than object or verb-phrase ‘only’545

(Crain et al. 1994), raising the possibility that this interpretation might be more546

difficult for adults to process. We think this is unlikely. Other developmental studies547

find that subject ‘only’ is no more difficult for children than verb-phrase ‘only’ and548

there is no evidence that we know of to suggest that adults have substantial difficulty549

with these forms (see Paterson et al. 2003; Paterson et al. 2006).550

A second difference between the present study and that of Kim and colleagues is551

whether the inference in question leads one to prefer the previously mentioned item552

or reject it. In Kim’s study, hearing ‘only’ leads the listener to construct a context set553

based on the previously mentioned items, encouraging looks to these referents. In554

our study, hearing ‘only’should lead participants to infer that a previously mentioned555

items cannot be the object of the target sentence. Notice that this inference involves556

implicit negation (exclude the previously mentioned items) and the need to avoid557

looking at objects that were previously relevant. Both of these things could make558

processing more difficult (Wason 1965, Carpenter and Just 1975, Dale and Duran559

2011 among others). One virtue of this account is that explains why we see a rapid560

effect in the ‘also’ condition: this inference, like Kim’s, involves a direct preference561

for previously mentioned items, rather than an implicit negation.562

Acknowledgement For conversations and inspiration on this project, we are very grateful to Gen-563

naro Chierchia, Josh Hartshorne, Hazel Pearson, and the audiences of Language and Cognition at564

Harvard University in 2008 and 2009. A special thanks goes to Clemens Mayr and Tim O’ Donnell,565

who contributed to the discussions that lead to these experiments. We wish to thank also Tracy566

Brookhyser, Carlyn Friedberg, Margarita Zeitlin, Nina Hrebenko and Miranda Fidler for assistance567

with data collection and preparation of the stimuli.568

5 Appendix A: Off-line Norming Study on Also and Only569

To ensure that inferences that we were studying were robust, we conducted an offline570

sentence completion study. The goal of this task was to verify that, given adequate571

time, participants could infer that the direct object in the ‘only’ sentence must be572

novel to the discourse, while the direct object in the ‘also’ sentences must have been573

mentioned in the context sentence. This experiment employed the materials from574

Experiment 1.575

Sixty-four participants were recruited through the on-line crowd sourcing tool576

Amazon Mechanical Turk. They received $ 0.50 for their participation. Twenty eight577

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



additional participants were excluded (26 for poor accuracy on the filler trials and578

2 because English was not their native language). The design was almost identical579

to Experiment 1. Instead of hearing instructions, participants read the introduction,580

context and target sentences, except that last word of the target sentence was truncated581

after the first letter, as in (27).582

(26) Jane also has some c__583

Participants were shown the display from Experiment 1 and clicked on the picture that584

completed the sentence. Note that control-old and control-new trials were identical585

in this study, since the final word which differentiates these conditions is omitted.586

There were 8 ‘also’ trials, 8 ‘only’ trials,16 control trials, and 8 fillers which were587

used to filter out inattentive participants.588

A logistic regression analysis was performed comparing ‘old-cohort’ choices for589

‘also’ and ‘only’ sentences to the control sentences. Both the ‘also’ (M = 73.7 %, z590

= 18.935, p < 0.001) and the ‘only’ (M = 15.8 %, z = 3.973, p < 0.001) responses591

were significantly different from the control responses (M = 23.9 %). Therefore, we592

confirmed that participants were able to make the relevant inference and select the593

intended image for both the ‘also’ and the ‘only’ sentences.594

6 Appendix B: Norming Study on the Intonation of Also595

When ‘also’appears between the subject and the verb it has two possible readings. We596

wanted our participants to get the reading in which ‘also’ associates with the subject.597

There is no other way to convey this reading in contemporary English. While the598

oldest co-author favored ‘John too has some candles,’her younger colleagues insisted599

that no one spoke like this anymore. So we were forced to use prosody to disambiguate600

the intended reading. We did this by producing the sentences with the prosody given601

in (28). The capitalization indicates prosodic focus on the subject and the commas602

signal a prosodic break before the verb.603

(27) JANE, also, has some candies.604

It was our intuition that when the sentence was produced in this way, the only possible605

reading is the one where the presupposition is on the subject (Jane, in addition to606

someone else, has some candies). However, to make sure that folks did not get607

the reading where the presupposition was on the verb phrase (Jane, in addition to608

something else, has some candies), we conducted a norming study.609

Ten undergraduate students participated in the experiment. The experiment was610

conducted using Psyscope X. The participants were been presented with 16 pre-611

recorded auditory stimuli like (28) above. These were a subset of the utterances that612

were used in the ‘also’ condition of Experiment 1. For each one they were asked to613

select the correct interpretation from two alternatives.614

(28) a. Jane has some candies and she has something else too.615

b. Jane has some candies and someone else has candies too.616

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


20 J. Romoli et al.

Participants made their selection by pressing a keyboard button corresponding to617

the side of the screen where the alternatives was presented. The position of the alter-618

natives on the screen was counterbalanced and the presentation order for the items619

was randomized for each subject. The stimuli were interspersed with 48 fillers. These620

included utterances with too as in (30) (followed by the same possible choices) and621

utterances like (31), (followed by a choice between ‘Justin is Christina’s neighbor’622

or ‘Justin is Christina’s neighbor and Christina is Justin’s neighbor’).623

(29) JANE, too, has some candies.624

(30) Justin and Christina are neighbors.625

Participants judged the critical sentences to have the subject-association reading 84 %626

of the time.AWilcoxon-Signed-RankTest showed that this was significantly different627

from chance (Z = 2.28, p < 0.05). Thus the intonation pattern on the critical sentences628

strongly biases participants to interpret ‘also’ as associating with the subject.629

References630

Abusch, D. 2010. Presupposition triggering from alternatives. Journal of Semantics 27 (1): 1–44.631

doi:10.1093/jos/ffp009632

Allopenna, P. D., J. S. Magnuson, and M. K. Tanenhaus. 1998. Tracking the time course of spoken633

word recognition using eye movements: Evidence for continuous mapping models. Journal of634

Memory and Language 38:419–439.635

Beaver, D. 2001. Presupposition and assertion in dynamic semantics. Stanford: CSLI Publications.636

Beaver, D., and B. Z. Clark. 2009. Sense and sensitivity. How focus determines meaning. Oxford:637

Oxford University Press.638

Beaver, D., and B. Geurts. To appear. Presuppositions. In Semantics: An International Handbook639

of Natural Language Meaning, vol. 3. ed. C. Maienborn, K. von Heusinger, and P. Portner.640

Berlin: Mouton de Gruyter.641

Bott, L., and I. Noveck. 2004. Some utterances are underinformative. Journal of Memory and642

Language 51:437–457.643

Carpenter, P., and M. Just. 1975. Sentence comprehension: A psycholinguistic processing model644

of verification. Psychological Review 82:45–73.645

Chemla, E. 2009. Presuppositions of quantified sentences: Experimental data. Natural Language646

Semantics 17 (4): 299–340. doi:10.1007/s11050-009-9043-9.647

Chemla, E., and L. Bott. 2013. Processing presuppositions: Dynamic semantics vs pragmatic648

enrichment. Language and Cognitive Processes 38 (3): 241–260.649

Chierchia, G., and S. McConnell-Ginet. 2000. Meaning and grammar: An introduction to semantics,650

2nd ed. Cambridge: MIT Press.651

Crain, S., W. Ni, and L. Conway. 1994. Learning, parsing, and modularity. In Perspective on652

sentence processing, ed. J. Clifton, Charles L. Frazier, and K. Rayner, 433–467. Hillsdale:653

Lawrence Erlbaum Associates.654

Dale, R., and N. D. Duran. 2011. The cognitive dynamics of negated sentence verification. Cognitive655

Science 35:983–996.656

Gazdar, G. 1979. Pragmatics: Implicature, presupposition, and logical form. New York: Academic657

Press.658

Heim, I. 1983. On the projection problem for presuppositions. In Proceedings of WCCFL 2, ed. D.659

P. Flickinger, 114–125. Stanford: CSLI Publications.660

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f



Heim, I. 1992. Presupposition projection and the semantics of attitude verbs. Journal of Semantics661

9:183–221.662

Huang, Y. T., and J. Snedeker. 2009. Online interpretation of scalar-quantifiers: Insight in the663

semantics-pragmatics interface. Cognitive Psychology 58 (3): 376–415.664

Karttunen, L. 1974. Presupposition and linguistic context. Theoretical Linguistics 1:181–194.665

Kim, C. 2007. Processing presupposition: Verifying sentences with ‘only’. 31st Penn Linguistics666

Colloquium.667

Kim, C., C. Gunlogson, M. Tanenhaus, and J. Runner. 2008. Focus alternatives and contextual668

domain restriction: A visual world eye-tracking study on the interpretation of ‘only’. Proceedings669

of SuB 13.670

Kripke, S. 2009. Presupposition and anaphora. Linguistic Inquiry 40 (3): 1–21.671

Matin, E., K. Shao, and K. Boff. 1993. Saccadic overhead: Information processing time with and672

without saccades. Perception & Psychophysics 53 (4): 372–380.673

Paterson, K., S. Liversedge, C. Rowland, and R. Filik. 2003. Children’s comprehension of sentences674

with focus particles. Cognition 89 (3): 263–294.675

Paterson, K., S. Liversedge, F. White, Diane, and K. Jaz. 2006. Children’s interpretation of676

ambiguous focus in sentences with ‘only’. Language Acquisition 13 (3): 253–284.677

Romoli, J. 2012. Soft but strong: Neg-raising, soft triggers, and exhaustification. Ph. D. thesis,678

Harvard University.679

Romoli, J. To appear. The presuppositions of soft triggers are obligatory scalar implicatures. Journal [AQ1]680of semantics.681

Romoli, J., and F. Schwarz. To appear. An experimental comparison between presuppositions and682

indirect scalar implicatures. In Experimental Perspectives on Presuppositions, ed. F. Schwarz.683

Studies in Theoretical Psycholinguistics. Dordrecht: Springer.684

Schlenker, P. 2008. Be articulate: A pragmatic theory of presupposition projection. Theoretical685

Linguistics 34 (3): 157–212.686

Schlenker, P. 2009. Local contexts. Semantics and Pragmatics 2 (3): 1–78.687

Schwarz, F. 2007. Processing presupposed content. Journal of Semantics 24 (4): 373–416.688

Schwarz, F., and S. Tiemann. 2013a. The path of presupposition projection in processing—the689

case of conditionals. In Proceedings of SuB 17, ed. E. Chemla, V. Homer, and G. Winterstein,690

pp. 509–526.691

Schwarz, F., and S. Tiemann. 2013b. Presupposition projection in online processing. (Ms.,692

submitted).693

Simons, M. 2001. On the conversational basis of some presuppositions. In Semantics and Linguistic694

Theory (SALT) 11, ed. R. Hastings, B. Jackson, and Z. Zvolenszky, 431–448.695

Stalnaker, R. 1974. Pragmatic presuppositions. In Semantics and Philosophy, ed. M. Munitz and696

D. Unger, 197–213. New York University Press. [AQ2]697Stalnaker, R. 1978. Assertion. Syntax and Semantics 9:315–332.698

Tanenhaus, M. K., M. J. Spivey-Knowlton, M. K. Eberhard, and J. Sedivy. 1995. Integration699

of visual and linguistic information in spoken language comprehension. Science 268 (5217):700

1632–1634.701

van der Sandt, R. 1992. Presupposition projection as anaphora resolution. Journal of Semantics702

9:333–377.703

Wason, P. 1965. The contexts of plausible denial. Journal of verbal learning and verbal behavior704

4:7–11.705

Aut

hor's

Pro

of

Unc

orre

cted

Proo

f


Author Query706

AQ1 Please provide the complete details for reference “Romoli (To appear)”.707

AQ2 Please provide the publisher name for reference “Stalnaker (1974)”.708

Aut

hor's

Pro

of

Resolving Temporary Referential Ambiguity Using Presupposed Content

Documents

Transcript of Resolving Temporary Referential Ambiguity Using Presupposed Content